CHAPTER 6

MAC LAYER PROTOCOLS

6.1 INTRODUCTION

Medium access control layer which deals with the allocation of radio channel

resources, plays a very important role for QoS guarantees in WATM networks [128].

Three MAC layer protocols, namely, Combined Hybrid ARQ and Adaptive scheduling,

RAPP power control algorithm and Distributed call admission control are presented in

this chapter. These protocols enhance the overall quality of WATM networks.

It is desired that a MAC protocol design should be coupled with an error control

method. The present work describes an improved combined hybrid ARQ error control

and adaptive scheduling scheme for time division multiple access/time division duplex

(TDMNTDD) MAC protocols in WATM networks. The improved algorithm has more

CRC bytes than the earlier one.

The main challenge in wireless communication is to overcome the disadvantage

of the hostile characteristics of the radio channel. Specifically, power control is one of

the major concerns during retransmission in wireless networks. The present work

explains a version of Retransmission Algorithm based on Power Priorities (RAPP) for

power control in WATM networks. The algorithm is implemented in a Rayleigh fading

environment.

In general, admission decisions in WATM networks are made to ensure

guaranteed QoS for heterogeneous, traffic [129]. In the present work, the improved

DCAC scheme is described for WATM networks in which resource reservation is

dwamically allocated according to the user mobility information to meet the QoS

requirements for heterogeneous traffic. Along with blocking and dropping probabilities,

admission delay is also analysed for the improved scheme in a Rayleigh fading

environment.

6.2 COMBINED HYBFUD ARQ AM) ADAPTIVE SCHEDULING

A combined hybrid automatic repeat request error control and adaptive

scheduling scheme is proposed here for time-division multiple access/time-division

duplex medium access control protocols in WATM networks. This scheme is

appropriate for supporting real-time traffic with strict QoS requirements. As the channel

BER in wireless environment is very high and varies corn to lo-', existing schemes

have been using FEC to overcome the noisy channel condition at the cost of valuable

bandwidth. The proposed scheme guarantees the throughput requested by a real-time

trafEc user while keeping the bandwidth consumption at a minimum for a TDMA

system.

6.2.1 System Model

The proposed scheme is based on a finite-state Markov channel model. In this

model, only the transitions of two adjacent states are allowed. The switching process

between states is described by the transition probability matrix.

Figure 6.1 Two-state Markov model

Figure 6.1 shows the two-state Markov channel model. In this model, the channel

state varies between two states namely, '0' (bad) and ' 1 ' (good), which could be fiuther

referred to as fading and non fading states respectively. The transition matrix of the

model can be given as,

The punctured convolutional codes are used with Viterbi algorithm to enable

adaptive encoding and decoding without modifring the basic structure of the encoder

and the decoder. The error correction is perfmed by shortened cyclic codes with

variable degrees of shortening. When the channel bit error rate increases, the system

generates additional parity bits for error control [130].

The adaptive error control algorithm is illustrated by the block diagram in Figure

6.2. Information is generated by the data source and encoded using one of the M

Punctured Convolutional coders. An interleaver is used to randomize the burst enor.

The forward channel is composed of the interleaver and the modulator. In order to

implement the adaptive coding scheme, it is necessary to use a return channel. It is

assumed that the return channel is enor-free.

Figure 6.2 Adaptive Error Control

t

De- Multiplexer

The Channel State Estimator (CSE) determines the current channel: state based on

counting the number of erroneous blocks, Once the channel state has been estimated, a

decision is to be made by CSE to change the code. Such changes and the corresponding

messages are to be sent to the encoder and the decoder. In FEC schemes, only error

correction is performed, while in hybrid ARQ schemes, retransmission of erroneous

blocks is requested whenever the decoded data is labeled as unreliable.

Return Channel

4 - Channel state ': Estimator

Data Source I-IIIIII-

Encoder --.* + - lnterleaver Forward De- Decoder

Channel --* lnterleaver 4 -

6.2.2 Medium Access Control Protocol in WATM

A MAC protocol is a set of rules to control the access of a shared wireless

communication medium among various users. These protocols need to provide access to

the users moving within the cell and handoff calls. At the same time, they must have

features to include ATM service standards. Multiple access schemes and protocols can

be static or dynamic. In general, multiple access schemes can be classified into three

main categories namely, Fixed assignment, Random access and Demand assignment.

Unlike fixed assignment, demand assignment schemes avoid wastage of

bandwidth by assigning it on demand. Demand assignment protocols also avoid

bandwidth wasted due to collision by providing the connections with free contention

bandwidth during active periods unlike random assignment. Because of these features,

MAC protocols based on demand assignment are most appropriate for the requirements

of integrated wireless networks.

Demand assignment users are required to provide information regarding their

needs for bandwidth which will be assigned on demand. It is widely accepted that the

dominant services in the broadband environment are VBR services. Therefore for VBR

users, bandwidth will be assigned according to requirement. Whenever the VBR user

enters an idle period, the assigned bandwidth will be allocated to another user f 13 1 I.

In demand assignment protocols, the channel bandwidth is time-slotted and

represented by a single or several block(s), depending on the channel rate and the type of

applications. Figure 6.3 shows the radio channel bandwidth allocation. Each block is

divided into uplink and downlink periods (channels), each of which is further divided

into sub periods. This type of classification might be performed on a slot-by-slot or

period basis. Downlink traffic can be transmitted in a separate channel (using a different

frequency band), or it can time share a single block with the uplink traffic by usingtime-

division duplex (TDD). The latter provides better flexibility in controlling the available

bandwidth by dynamically allocating the length of each period.

The uplink period is a mobile unit-to-BS transmission, subdivided into request

access (RA) and data transmission access (TA) sub periods. In the RA sub period, users

transmit their RA packets to the BS for requesting bandwidth. The BS will identify the

successful users (those not involved in collision) and assign them a bandwidth, if

available.

Figure 6 3 Radio channel bandwidth allocation in WATM

Down Link Up link

The time taken for the transmission from BS to mobile unit is the downlink

period. The downlink period is subdivided into acknowledgment (ACK) and data

downstream (DD) sub periods, Basically, in the ACK sub period, the BS notifies the RA

terminals about their request status. In the DD sub period, the BS transmits downstream

data to the destined terminals. Downlink transmissions are governed by the BS and

generally supported by a contention-f?ee time-division multiplexing (TDM) broadcast

mode. Since a TDM channel can be operated with high efficiency and low delay, the

downlink channel is not a critical performance driver of the system.

4 L A x b

ACK C T DD Ft-- R A d I TA

I I I

....a ...*(I

As explained, the uplink period is subdivided into RA and TA sub periods. It is

to be noted that the number of terminals under BS coverage is considerred to be much

larger than the available channel bandwidth. However, not all the terminals are active

simultaneously. In this case, the MAC scheme used in the RA sub period is random.

I * Scheduler I

Power of the received signal

Queue length information

L i

State of the mobile user

t I Slot allocation for the user

I

I Mobile Terminal

Figure 6.4 Scheduler

In general, demand assignment protocols provide better channel utilization than

fixed assignment and random assignment schemes. Because of the frequent bandwidth

assignment, there will be an overhead due to channel access delay and control signal

exchanges, which might lead to inefficiency. The channel access delay is classified into

connection and packet access delay. The channel access delay is defined as the time

required for the RA and TA packets to access the channel. The connection access delay

is experienced in the RA sub channel and will have a significant impact on the

retransmission requests. The packet access delay can be defined as the time required for

active VBR, ABR, and UBR applications to resume transmission after getting back fiom

an idle period. The packet access delay is experienced in the TA sub channel. This

problem has a great impact on the queue length at the terminal premises. For real-time

active VBR services, packets will be stored at the output queue at the user terminals until

&ey get another reservation for transmission after getting back from an idle period.

Therefore, the output queue might go off before the bandwidth reservation is p t d .

For non-real-time active ABR services, data might be stored at the higher layers;

therefore, packet access delay would not affect the terminal's output queues.

The proposed MAC protocol operates based on time slots. When a user is in the

noisy channel, the system assigns an extra slot for that user, which is twice as many slots

as needed for the user in the quiet channel. The extra slot is assigned to provide QoS

guarantees for the user in the noisy environment. Considering the small size of a cell

covered by a base station, the propagation delay between the base station and a mobile

user can be assumed as being negligible.

UPSTREAM STRUCTURE Queue status slots

Signaling slots used for feedback M message slots 4 b

DOWNSTREAM STRUCTURE

M message slots Allocation slots Signaling slots 44-•

Figure 6.5 Upstream and Downstream Structures

The scheduler shown in Figure 6.4 has complete information about channel state

information (CSI). Figure 6.5 shows the general structure for upstream and downstream.

The upstream structure consists of signaling mini slots for the channel access that uses

the slotted ALOHA protocol. Active mobile users use queue status mini slots to

feedback their queue information to the base station. Based on the information of the

queue status and the channel state, the scheduling algorithm allocates the appropriate

number af slots to guarantee the QoS requested by active users. The number of queue

status mini slots is equal to the maximum number of mobile users that can be served

within a block. The message slots are used for the information transmission. The

scheduler needs to reserve certain number of slots for active users possibly in the noisy

state. The reserved slots are used for users in the noisy channel to meet their QoS

requirements. If some slots are still left unassigned, they will be allocated to new calls.

6.2.3 Performance Analysis of Combined Hybrid ARQ and Adaptive Scheduling

The main function of a hybrid ARQ protocol is to properly combine ARQ and

FEC techniques. There are two types of hybrid ARQ schemes namely, type I and type I1

hybrid ARQ. In type I hybrid ARQ, each packet is encoded for both error-detection a d

error-correction capabilities. Retransmission is initiated once the error correction fails as

indicated by the error detection to recover all the errors incurred. In type-11 hybrid-ARQ

protocol, a code word is encoded using two different codes, C1 and C2. These codes have

corresponding decoding operations Dl and D2 which can recover the original code word

from noise corrupted versions of C1 and Cz respectively. CI is initially transmitted while

Cz is set aside. Upon receiving C1, the receiver attempts decoding operation Dl to

recover the transmitted data. If the attempt is successfil, the receiver sends an

aclcnowledgrnent (ACK) to the transmitter; otherwise, a negative acknowledgment

(NACK) is sent. The transmitter responds to the NACK by sending C2. The receiver then

attempts to decode Cz, using decoding operation D2. If successfil, the receiver finds the

corrected version of C2 to recover the desired infomation and sends an ACK to the

receiver. If unswccessfil, the receiver combines C1 and C2 to create C3, a new code word

with lower rate. If the third decoding operation D3 is succe~sful, the data is recovered

and an ACK is sent to the receiver; otherwise, a NACK is sent and the entire process is

repeated.

In this work, a punctured RS code with a finite Galois Field GF (2' ) is

considered. Specifically, a type I1 hybrid ARQ has been adopted. The throughput of type

I and type II hybrid ARQs with the developed two-state Markov channel model are

estimated. White transmitter and receiver buffers and noiseless feedback channel with

selective-repeat protocol have been assumed in the ARQ retransmission.

The overall throughput for the two-state Markov model shown in Figure 6.1 can

be expressed as,

q " P o ~ o + P ~ 91 (6.2)

where, pi (i=0,1) and qi are the steady state probability and throughput of state i

respectively.

The throughputs Ti (i=O, 1) of type I1 hybrid ARQ can be analyzed first. Let Tsi

denote the total number of transmissions (including the original transmission) required

for a packet to be decoded successfblly in state i(i=O,l). Then the throughput l\i (i=O,l)

can be expressed as,

where, E[Tsi ] is the average number of transmissions and Wn is the coding rate of

codes Gland C2. Figure 6.6 shows a state diagram for the behavior of a type I1 hybrid

ARQ decoder. In this figure, Dl, D:! and D3 denote the decoding operations for codes CI,

C2, and C3 respectively and PR/') , P$-) , pR13) are the probabilities of the generation of a

retransmission request by Dl, Dz and D3 respectively. Now E[Tsj] may be computed as

follows:

where,

1 k k Then, q = - - 1 - P ) . i =0,1

E[T',I TI n

As the characteristics of the wireless channel vary with time, it is desirable that

the error control method and the scheduling algorithm be adaptive according to the

varying channel conditions.

The scheduling scheme in the WATM MAC protocol must alIow isolation and

integration of the different service categories. Fair-queuing or General Processor Sharing

(GPS) strategies, especially Self Clocked Fair-Queuing (SCFQ), have to provide such

properties. The basic idea of SCFQ scheduling is that packets are served based on their

service tags which are assigned to packets in the head-of-line position of a queue,

In particular, the service tags can be calculated by the following equation:

=i Fi = - + max (4 , ~ ( a i ) ) Pi

where,

Fi is the service tag of the im packet

Li is the length of the i packet

ai is the anival time of the i& packet

v(t) is the service tag of the packet in service at time t ;

Fi is the weight of connection i.

In wireless environments, the bandwidth requirement of a real-time connection

varies according to the channel condition. In WATM networks, adaptive SCFQ is

particularly desirable for the case of bandwidth assignment of retransmission. In this

work, SCFQ scheduling with adaptive weights for a wireless ATM MAC protocol is

presented. Specifically, the service weight of each connection dynamically changes

according to the channel condition as well as its QoS requirements.

Figure 6.6 State diagram of a Type-I1 Hybrid ARQ protocol

It has been shown that GPS is able to make worst case performance guarantees

on throughput and delay by the use of leaky bucket rate control. In particular, suppose

there are Z connections sharing one link with the capacity of Ch and that connection 'i' is

assigned with a service weight. Then, the given connection 'i ' guaranteed with a service

rate is given by,

xi'' g, =- ' X i

A x j

Accordingly, the service weight for a constant bit rate connection ( X ) can easily

be determined by its peak cell rate. For an available bit rate connection, its service

weight could be equal to minimum cell rate. Likewise, in order to provide at least a

minimum QoS to unspecified bit rate (UBR) connections, some bandwidth may be

reserved but with a small service rate. On the other hand, the weight determination of a

variable bit rate (VBR) connection is not straightforward. Suppose a VBR connection is leaky bucket policed then its service weight should be associated with its traffic

parameters, such as peak cell rate, sustained cell rate, burst tolerance, maximum transfer

delay, etc. For example, the VBR source is characterized by (m, B,, Bp, Dp ), where rn is

the maximum burstiness of the source, B, is the average cell rate of the source, B, is the

peak cell rate of the source and D, the maximum tolerable delay of the source. The

regulation of the leaky bucket is characterized by (a, p ), where @ is the token buffer

size and p is the token generation rate. WATM cells that have passed the leaky bucket

are stored in the ready to transmit (RTT) buffer and will be served at the prescribed rate.

Then the delay of a packet is found to be the difference between the time it reaches the

other side over the wireless channel and the time it arrives at the RTT buffer.

The maximum burst size of a leaky bucket policed VBR source is given by,

Then, under the assumption that gi 2 p 2 B,, the maximum queue length of the RTT

buffer is,

Moreover, if the request and reservation delay are ignored, then the maximum delay is

given by,

Once this delay requirement is met, the guaranteed service rate can be determined as,

Finally, the service weight of the VBR source can be obtained as,

In a non fading duration, each packet most likely needs only one transmission.

~ u t more transmissions might be necessary in the case of fading duration. As a result,

the service weight of real-time connection should be increased by a factor of two during

fading. On the other hand, the service weight of a non-real-time connection during

fading will be reduced to one-half or one-fourth (or even less) of the one without fading.

Specifically, the combination of adaptive fair-queuing and hybrid type I1 ARQ scheme

for a connection can be illustrated by the flowchart shown in Figure 6.7, where the

channel state can be determined by whether it can decode the information bytes correctly

from the received packet by C1 or C2 code. Simulations presented in the next section will

show that this scheme is quite reasonable in terms of improved throughput and delay

performance of the overall system.

In simulation, the channel capacity is assumed as 1 Mb/s, and a variable block

length is employed with maximum number of 20 data slots, including the block header

and mini slots. The transceiver turn around time is ignored but the guard time and

physical layer preamble are taken into account. The Rayleigh fading channel is

considered for simulation. Infinite transmitter and receiver buffers and noiseless

feedback channel are assumed. For service traffic, it is assumed that six real-time VBR

(rtVBR) and six UBR connections exist in the system. The rtVBR and UBR connections

are modeled as on-off and Poisson sources, respectively. Accordingly, the traf%~c load of

the system is about 70% (18% rtVBR traMic and 52% UBR traffic).

Figure 6.8 shows the system throughput of different retransmission and

scheduling schemes in terns of different coding rate (R=k/n) of C1 code. Here the

WATM packet size is 55 bytes.

It is observed that the system reaches the maximum throughput when R=l

( k=55,n=55 ) for both type I and type I1 hybrid ARQ. More importantly, the combined

Hybrid type I1 ARQ and adaptive scheduling yields the best performance especially with low coding rate.

A A I, 7

BS performs traffic scheddmg based on current weight of each connection

1 BS sends allocation domat ion and

ACKs of dink traffic to MTs

I A

f BS receives the packets sent by hiTs and the ACHs For

the downlink traffic

Set charmel state flag = good Weight of VBR connection is

divided by 2 Weight of UBR comection

is muhplied by 4

Figure 6.7 Flowchart for the Hybrid Type I1 ARQ and Adaptive Scheduling scheme

BS transmits downlink traffic bared on hybrid type-11 ARQ

Set charmel state flag= bad We$t of V B R connection

A

1 Each mobile transmits packets 1

I

based on type-I1 hybrid ARQ and ACK for do& traffic

is muhiplied by 2 Weight of UBR connechon

is dxoided by 4

41

-4- type- I1 hybrid ARQ

0.95

coding rates R=k/n k=55

Figure 6.8 Throughput of different retransmission schemes

Figure 6.9 shows the average delay of UBR traffic with respect to different

coding rates. Simulation results indicate that the average delay of UBR traffic for the

case of type I hybrid ARQ is minimized when the coding rate R=0.85, while the delays

for the case of type 11 hybrid ARQ and type II hybrid ARQ with adaptive fair-queuing

are both minimized when the coding rate R=l. This clearly shows that type I hybrid

ARQ always has a larger delay.

Figure 6.10 shows the mean delay of UBR traffic as a fkction of traffic load

which is increased by the increase of the mean rate of Ul3R traflic. It can be observed

that the UBR traffic has the largest mean delay in the case of type I ARQ. The mean

delay of UBR traffic can be significantly reduced by the use of type I1 hybrid ARQ and

M e r reduced by using the combination of type I . hybrid ARQ and adaptive fair-

queueing, which has the best performance, especially under a heavy load condition.

Figure 6.9 Average delay of UBR Traffic Vs coding rate

TmRic Load

Figure 6.10 Average delay of USIR services Vs Traffic load

70

60

- ga- g

40 In 3 C

" - - Q) w

$20

10

8s

I 1 a 1 - type-l hybrid ARQ 4- type-ll hybrid ARQ <,

- - type-ll hybrid ARQaadaptive fair-queueing -

-

-

- /

b - $ " .m

0.65 0.7 0.75 0.8 0.85

0.09 , , , , , , , / - type-1 hybrid ARQ

0.08 - -+- type-ll hybrid ARQ - type-ll hybrid ARQ&adaptiva fair-queueing -

0.07 -

Traffic Load

Figure 6.11 VBR cell loss rate Vs Traffic load

Figure 6.11 shows the cell loss rate (CLR) of VBR traffic and average delay of

UBR traffic in terms of traffic load which is increased by increasing the source rate of

the VBR traffic. By the use of a combination of type I1 hybrid ARQ and adaptive fair-

queuing, the CLR of VBR tr&c can be kept under a low value. It is M e r noticed that

in some traffic load regions, the type I1 ARQ without adaptive fair-queuing has better

UBR delay performance than type I1 hybrid ARQ with adaptive fair-queuing, as the

VBR traffic suffers a larger CLR.

6.3 RETRANSMISSION ALGORITHM BASED ON POWER PRIOTUTIES

The main challenge in wireless communication is to overcome the disadvantage

of the hostile characteristics presented by the radio channel. In this regard, power control

is one of the important issues in the environment of retransmission. The proposed

Retransmission Algorithms based on Power Priorities (RAPP). may be used for power

control at the retransmission of request packets sent by remote terminals to the base

station @S) in WATM networks.

63.1 Priorities using different power levels

The proposed power control scheme considers a TDMA / TDD access scheme

with fixed block size of 64 slots. As part of this framework, a scheduling mechanism that

assigns slots to the terminals in a block-by-block basis has been developed for the base

station. It allows and promotes the satisfaction of QoS guarantees for voice, data and

video services. The scheduler handles the five standardized types of services (CBR, rt-

VBR, nrt-VBR., ABR, and UBR), assigning priorities according to Table 6.1, where a

bigger number implies a higher priority.

Table 6.1 Priorities for the different types of senices in WATM

The idea behind RAPP is to help the WATM cells that have failed to reach the

receiver on the request procedure. After each failure, there are a set of power levels from

which a contending terminal may randomly choose increased power until the

predetermined limit. Providing a terminal the opportunity to use a higher transmission

power increases its capture probability at the receiver. At the same time, by potentially

increasing the power level after each retransmission, the f h e s s in serving the earlier

requesting terminals first is assured to some extent.

Type of service

Priority

An important motivation to use the a4PP algorithm is to take advantage of the

capture effect which has been shown to increase the throughput. Using different power

levels has the advantage as compared to back off mechanisms in that no waiting periods

are necessary and still one af the terminals has a very good chance of being successful.

Further capture effect is considered in the receiver's side. QoS restrictions are imposed

by the WATM network such as the need to guarantee a maximum. delay, cell loss ratia

and to limit the maximum number of retransmissions wbich is very important for delay

sensitive services.

CBR

5

rt-VBR

4

nrt-VBR

3

ABR

2

UBR

1

The following rule is proposed to specify the different power levels

,*...--, pm) to be used in RAPP algorithm so that contenders who have waited for

longer time still have an advantage over new ones, regardless of the total number of

contenders and the power levels they are using,

~=ZIG; for z 1 5 i 5 m

where 'm' is the number of power levels given by,

m= log,,(^, +f l ) i -1

This results in power vector as follows:

[P,, P, . ......... , Pm ] = IP, . z.P, , z2..p2 ,......... .., zm -'.P, 1 (6.18)

As an example, it is assumed that, PI = 0. 625 mW , P, = 160 Mw and z = 2 to

capture the power vector generated. The m,, power levels are determined according to

the minimum (PL) and maximum (P,) allowed transmission power levels for the

terminals. The minimum power (PL) must satisfy SNR and bit error rate PER)

constraints. Assuming a given PL and P,, the minimum power level to be used by a

terminal PI should be above PL, while the maximum power level P, should be below P,, .

63.2 RAPP Algorithm

Two priorities are defined in this protocol; the higher priority is for real time

applications (CBFt, and rt-VBR) and the lower priority is for non real time applications

(nrt-VBR, ABR, and UBR). The set of power levels is divided into two subsets, each

corresponding to one of the priorities. Let the subsets be defined as: (PI, Pz,. . . . . ..Pk) for

low priority and (Pk+il, Pk+2,. . . .., Pm) for high priority tamhds.

In its first attempt, each terminal will send its request through one of the request

slot using the lowest possible power level corresponding to its priority. After each

collision, the terminal will select a new power level from among those allowed by the

R,APP protocol according to its priority level and the number of failed attempts. The

operation of the retransmission algorithm is as follows:

If the mobile terminal (MT) corresponds to a high priority service, it will select

between Pk+l and Pk+2 in its first retransmission. After each new collision, it

will add to the allowed set of power level that is right above the highest one the

set currently has. When there are no more power leveIs to add, the MT will limit

itself to using one of the highest power levels and P,). If the MT fails to

reach the base station (BS) again, the process is stopped for a random time

selected fiom the set {T, 2T), where T is block duration. After this silence, if the

packet has not expired, the station will start trying again, choosing between the

highest power levels. If transmission is again unsuccessfil after a predetermined

number of attempts (m-k), it will stop again for a period selected from the set {T,

2T, 4T) and so on.

On the other hand, if the MT corresponds to low priority service, it will select

between PI and P2 in its first trammission. After each new collision, it will shift

the allowed set upwards but still keeping two options in the set, W e n it is not

possible to move upwards, i.e., when Pk is already a member of the allowable

set, the MT will transmit one for the last time using Pk and then, if the

transmission is again unsuccessful, it will stop sending requests for a random

time. When this happens, a M l cycle will have been completed. The process is

restarted later, repeating the procedure &om the fmt transmission.

The above procedure is shown in Figure 6.12. In all cases, regardless of the

priority the MT belongs to, it will randomly select its power level considering that all the

options in the allowable set are equally possible.

It is to be noted that there is no differentiation among the Pvrrs that correspond to

the same priority group. Therefore CBR and rt-VBR users share the same privileges.

However, RAF'P has to do with the delivery of request to the base station only. It is the

responsibility of the scheduler at the base station to make sure that the five priorities

shown in Table 6.1 are respected when it comes to actually allocating the bandwidth.

Figure 6.12 Power levels for RAPP algorithm

6.33 Performance Analysis of RAPP

The achievable throughput S in a slotted ALOHA environment under traffic

generated according to a Poisson process with average arrival rate G is given by,

When capture effect is taken into account,

" In s=h~-'(l+C- P [ ~ q ~ u r e / C n ] ] ,,, n!

In the Eq.(6.20), C, is used to denote collision with n contenders. The capture

probability is calculated for rnultipath fading distribution. Finally, when several mini-

slots (m) are provided and the users are allowed to choose randomly among them, the

total load h is evenly divided among the mini-slots, which results in the following

expression for the throughput per mini-slot:

Load

Figure 6.13 Throughput Vs Traffic load under RAPP

The relationship between throughput and traffic load is shown in Figure 6.13.

The generated traffic is the combined traffic, created by the five types of traffic

generators plus retransmissions. Here the throughput (q) is the measure of the effective

utilization of the request channels. The maximum throughput corresponds to the

r~eption at the base station of 3 or 7 successful request packets per frame. As already

mentioned, the TDMA access scheme is considered. From Figure 6.13, the throughput

achieved with RAPP is a maximum value of 0.65 which means that it corresponds to

around 4.5 out of 7 possible packets being correctly received per hme, The simulations

are carried out for different number of available mini slots (m=3, 6, 7). The performance

of a Retransmission algorithm based on power priorities has been analysed. Simulation

results show that the throughput achieved is greater in this scheme.

6.4 DECENTRALISED CALL ADMISSION CONTROL SCHEME

It is well known that ATM is a connection-oriented service. Before a user starts

transmitting over an ATM network, a connection has to be established. This is done at

the call set-up phase. The main objective of this set-up procedure is to establish a path

between the sender and the receiver which may involve one or more ATM switches. At

each of these ATM switches, resources have to be allocated to the new connection. An important part of the connection establishment procedure is the Call Admission Control

(CAC) function. CAC is defined as the set of actions taken by the network during call

set-up phase to determine whether a connection request can be accepted or rejected[132].

In this work, a Decentralised Call Admission Control (DCAC) scheme is

presented for WATM networks which statistically multiplex user mobility information

to estimate future demand and available resources. Resource reservation for handoff

calIs is dynamically adjusted according to the user mobility information to meet the QoS

requirements for heterogeneous traffic. The proposed decentralized CAC is an online

scheme which is executed partially on serving Access Point (AP) and partially at

Mobility Supporting (MS) ATM switch.

6.4-1 Estimation for new connections

To estimate the bandwidth, a mechanism is required to gather cell counts at each

connection and update the traffic disb5bution at the MS-ATM switch, Therefore each

switch is provided with a cell-count request timer that is associated with it. The cell-

count request timer starts when the switch is engaged. When a cell-count of switch j

expires, a cell-count request is issued to the switch for each connection that has node j as

its destination. A queue is used to keep track of the number of cell-count requests issued

(XI, X2.. .Xn) and the number of responses received. When enough cell-count samples

are collected, the effective bandwidth of the connection can k calculated using the

following equations:

Effective Bandwidth of the connection = hr (6) 6

where 6 is jitter bound given by,

The arrival rate is given by,

where,

Ti is the interval between cell counts

r, is the cell-Ioss ratio

b is the buffer space

X; is the cell-count requests issued

n is the number of cell-count samples

k~ is the asrival rate

When a new connection is received at the AP, the AP will attempt to reserve the

bandwidth for the connection ftom the Radio Resource Manager (RRM). The amount of

bandwidth required by a new connection which is referred to as the equivalent capacity

can be obtained in the following way. If a new connection with the parameters (R, br)

inputs a link with buffer capacity 'X', where 'R' is the peak bit rate of the connection

and Ytl' is the average duration of the connection active period (mean burst of the

the equivalent capacity C of that connection would be,

C=(Ra(l-p)-Xc ) + 'I ( a b l ( l - p ) - ~ ~ ) ~ + 4 ~ c p a tt(1-p)+(2atl(l-p)) (6.24)

where,

R is the peak bit rate of the connection

a is the smoothing co-eficient

tl is the average duration of the connection

p is the source utilization

& is the buffer capacity

The equivalent capacity in Eq.(6.24) is used as an estimate of the bandwidth

requested by the new connection which is then compared with the effective bandwidth of

the destination port calculated in Eq.(6.22). The result available is then used for the next

'update period' until the next update of available bandwidth is canied out. There is a

possibility that the connection might be accepted by the proposed method but later

rejected by the actual CAC as this method is the estimation.

6.46 CAC with user mobility information

The proposed CAC is focused on CBR and VBR traffic.. For ABR traffic, all new

and handoff requests are treated as the same. There is no resource reservation for ABR

handoff call. All existing ABR calls will share the remaining resources after serving

CBR and VBR calls fairly and efficientIy. Three aspects are included in the proposed

CAC scheme:

i) Resources reservation for potential handoff calls updated periodically at each

BS.

ii) New CAC performed whenever a BS receives a new call request.

iii) Handoff CAC performed whenever a handoff event occurs.

To admit a call (new or handoff), two conditions must be satisfied. Guaranteed

QoS for a particular call should be provided and QoS performance of all existing calls

should still be guaranteed after the admission. The same conditions are applied to

reserve resources for any CBR or VBR class.

The available time is divided into equal intervals of length 'z' and the time

sequence is given by .t= O,z, 22.. ., so that there is at the most one handoff event for any

active mobile terminal 'x' in each of the intervals. It is assumed that at the beginning of

each 'z' interval, the handoff probabilities, Pab (t), of each mobile k' from cell 'al' to

cell 'az' are calculated. The handoff probabilities are known to the mobile's current cell

'al' and the probabilistic information can be exchanged to all neighboring cells in order

to reserve necessary resources for handoff calls.

Let cell 'a,' be the reference cell and ni, and nj, are defined as the number of

equivalent handoff calls in the i& CBR class and jfh VBR calls respectively, which need

cell 'al' to reserve resources in the next 'T' interval. The number of equivalent handoff

calls for a imffic class is the average number of calls in the class. This will handoff to

cell 'al' in the next 'z' interval and can be calculated as the accumulated haudoff

probability of all the calls in the class that will handoff from the neighboring cells to cell

'al' in the next z interval.

At the beginning of every 'r' interval, resource reservation is updated at each BS

for all classes of the traffic fiom the highest (the first CBR class) to the lowest (the last

VBR class) priorities. The algorithm starts with CBR for bandwidth reservation. If the

current calculated class is CBR class, it accordingly calculates the jitter bound. If there is

no CBR class then it will reserve for VBR and determines the delay bound and QoS

which must be satisfied for any class.

In the proposed model, an issue of paramount impowce is the estimation of call

blocking probability (PB) and handoff dropping probability (Phd). As long as real time

measurements not available, the parameters can be seIected based on the average

speed of mobiles, their movement pattans and the coverage area of wireless cells. Once

the network is deployed and traflic measurements are available, these probabilities can

be refined based on the statistical analysis of such measurements. For example, the data

regarding the movement and holding time can be collected based on the statistical

analysis of the data. Then Pg and Phd can be defined accordingly. Here the location,

time of the day and other factors can be taken into consideration. An example of this

would be to collect mobility details and call holding time statistics in highway or street

micro cells, and define PB and Phd based on the time of the day to capture the peak hour

traffic characteristics and incorporate that in the call admission algorithm.

In fact these algorithms can be readily incorporated in existing wireless and

cellular networks. Each base station co~municates the number of wireless users in its

domain to all its neighboring base stations fiom the Mobile Switching Centre (MSC) in a

periodic manner. As new sonnation becomes available, the call admission controller of

a base station recalculates the admission threshold to reflect the change in the system.

Finally the rate of the change in the number of users in neighboring cells can be further

used to refine the movement parameters (PB and Phd) in order to increase the

performance of the call admission algorithm defined in terms of wireless utilization.

6.43 Performance AnaIysis of DCAC

The performance analysis of the proposed Decentdised Call Admission Control

scheme includes delay analysis and call rate analysis.

The WATM processor control system operates on a delayed basis when requests

are queued or scanned at regular intervals. There are 2 levels of control namely, level 1

and level 2. At level I, a processor control is involved in scanning and interfkcing with

customers. At level 2, central processors are involved when. all the data for a call request

are received. The delay in the procwor can be calculated as,

Average call delay = 1 1 r 2 1 1 - a )

where,

I /p is the holding time

p is the average call duration

a is the traffic load

It is extremely difficult to accurately estimate how a system will perform under

red time M i c conditions. For that, a traffrc simulation can be used in which the total

capacity is determined by the probabilities of false traffic and peak traffic. The total

capacity is evaluated for the entire call processing. The channel holding time is defined

as the time during which a new or handoff call occupies a radio channel in a given cell

and it is dependent on the mobility of the user. While this is similar to the call holding

time in the fixed telephone network, it is often a fraction of the total call duration in a

wireless cellular network. Most research work on CAC and bandwidth reservation

assumes that the channel holding times in all cells are independent and identically

distributed according to an exponential distribution [133). With the help of this total

capacity and holding time, admission delay is calculated and used in the simulation.

Call blocking rate is defined as the rate at which the caIls are blocked due to the

unavailability of resources. The actual blocking probability data must be used to check

the outcome from the Erlang B model. Although the difference can be up to 15 percent,

the Erlang B model is still considered as a good model for obtaining usefbl estimates.

The counting of a dropped call is after the call is established but before it is

properly terminated as described below. The definition of the phrase '%dl is established"'

means that the call is setup completely by the setup channel. If there is a possibiliw of a

call drop due to no available voice channels, this is counted as a blocked call, not a

dropped call. Call drops are caused by factors such as:

UnsuccessfUl completion Erom set-up channel to voice channel

e Blocking of handoffs (switching capacity)

e Unsuccessfbl handoffs (processor delay)

Interference (foreign source)

a Improper setting of system parameters

The percentage of call drops (Pcd) is expressed as the ratio of number of call

drops before completion to total number of accepted calls handled by set-up channels.

Sometimes call drops due to weak signals are called lost signals. This case happens

when the mobile or portable units are at a standstill and the radio carrier is changed from

a strong setup channel to a weak voice channel due to the selective frequency fading

phenomenon. The simulation of call blocking rate and call dropping rate is performed

for DCAC and compared with centralised CAC.

Consider a wireless mobile ATM network where several BSs are connected to an

ATM backbone network through a Mobile Switching Center (MSC). The multiple

access mode used is Time-Division Multiple Access (TDMA). Each BS can support

different types of trafjtic. This work is confined to the service area of a particular MSC.

The wireless system is characterized by a broadcast channel in the forward direction (or

downlink), where the BS is capable of broadcasting packets to all users and a muItiple

access channel in the reverse direction (or uplink), where all users are capable of sending

packets to the BS. The measures of performance are the call blocking probability

(P~)and handoff dropping probability (Phd). Call blocking and handoff dropping

probabilities are simulated and comparison is made with the centraiised scheme. This is

extended to handle heterogeneous services and its performance is also analysed. TO

achieve a complete view of the results, the following two scenarios are considered:

(a) A Centralised CAC method that uses the MS-ATM switch to select the route and

also to decide whether to accept the connection or not.

(b) A Decentdised. online CAC that uses the estimated available bandwidth from

the last update to decide whether to accept the connection or reject it.

There is a possibility that a connection rnight be accepted by the proposed

method but Iater rejected by the actual CAC because this methad is an estimation. The

main objective of the generic CAC introduced by the ATM F o m is to make a quick

decision on whether to execute the real CAC.

One dimensional (ID) cellular arrays are considered for simulation purpose. The

simulated system has five radio cells arranged in a circle to avoid the boundary effect.

Mobile unit can be in any one of the cells with equal probability. Let the initial location

where mobile unit originates its cdl request be uniformly distributed in the ID region of

its current cell, where D is equal to 1500 m. The mobile can move in either direction of

the 1D region with equal probability, and its velocity is uniformly distributed between 5

and 20 d s . The interval used to update the estimation is z = 60s.

The traffic classes in the system are CBR, VBR and ABR. It is assumed that the

call arrival process follows a Poisson distribution with parameters A, where l /h is varied

within the range of (0-30) sec and call duration of 1/p =50 sec. For any particular call, it

can be any of the three traffic classes. In the simulation, the call blocking rate and

handoff dropping rate for three traffic classes are determined by using the performance

bounds.

Figure 6.14 shows the difference between the estimated bandwidth and the

available bandwidth. It illustrates that the proposed method works accurately for a

network with heavy load conditions. If the update of the available bandwidth is done at

shorter intervals, then this estimation will be very dose to the accurate values.

Furthermore, if the network being examined does experience abrupt changes in traffic

loads, then the update period can be reduced even more to meet the network needs.

The decentralised and centdised methods are compared for the number of

connections admitted by each of them during fixed periods. Figure 6.15 shows clearly

how the proposed CAC scheme admits more connections when compared to other

methods. The connections are admitted on the basis of QoS. A call is admitted only

when it satisfies the predefined quality of service.

Figure 6.16 shows the number of connections admitted for the centralized CAC

and the proposed decentralized CAC. Figure 6.17 shows the considerable delays

resulting from the centralised to the proposed online decentralised method. Assume that

three APs have forwarded connection requests to the MS-ATM switch at the same time.

If the switch is to use the centralised approach, then the time required to take the

admission decision d l increase rapidly in the centralised approach in comparison to a

slight increase for the decentralised approach as shown in Figure 6.17.

Decentralised CAC admits more connections than centralized CAC. This is

shown in terms of the throughput. The result in Figure 6.18 also shows the delay in the

admission decision as the number of channels in each AP increases for both approaches.

It is clear that there is a reduction in the admission response times in the decentralised

approach compared to the centralised approach.

The call blocking and call dropping probabilities for centralised and distributed

CAC schemes are simulated and compared. At lower load conditions, the centralized

CAC system has a higher blocking probability than the decentralised call admission

control algorithm. By increasing the system capacity, the performance of both systems

becomes comparable at low load conditions. However, at high load conditions, the

decentralised CAC provides better performance. The results in Figure 6.19 and Figure

6.20 show that by increasing the system capacity, the performance of the DCAC is

eficient.

The calls are admitted only when it satisfies their corresponding bounds. Based

on that analysis, the blocking and dropping are simulated as shown in Figures 6.21 and 6.22. The CAC scheme that uses jitter bounds can reduce the forced handoff dropping

rate of both CBR and VBR dls , The CAC scheme using user mobility information can

achieve a much lower call blocking rate and higher resource utilization, while keeping

the handoff dropping rate at a very low level. Results show that the proposed scheme

achieves a much lower handoff call dropping rate than the centrdisd CAC.

Figure 6.14 Estimation of Bandwidth

Fipre 6.15 Number of connections admitted for the two methods

I

20 30 40 50 60 70 80 90 100 No.of BSs

Figure 6.16 Comparison of the Admission delay

Figure 6.17 Holding time Vs Admission delay

No.of Time slots

Figure 6.18 Comparison of the Throughput

- ,

I

$-

40; I - - 20 40 & - - - - 80 100

Mang bad per cell

Figure 6.19 Comparison of Blocking probability

Figure 6.20 Comparison of Dropping probability

Call Arrival hteml(s)

Figure 631 Call blocking probability for heterogeneous services in DCAC

---- 1 -*- CBR , I

- * VBR 1 / i-- ABR - i - _- - _ < ~

i *+.- 3-- -+- - - - - - - * ...i !

-- ol-.lk% I ! r ' I I ' I I ' :

0 5 10 15 20 25 30 Call Arrival interval(s)

Figure 6.22 Dropping probabilities for heterogeneous services in DCAC

6.5 SUMMARY

Three MAC layer protocols have been discussed in this chapter namely,

Combined Hybrid ARQ and Adaptive scheduling, RAPP power control algorithm and

Distributed caIl admission control. Results show that these protocols enhance the overall

quality of WATM networks. The results of these schemes are summarized as follows:

A combined hybrid ARQ error-control and adaptive scheduling scheme is

presented for WATM networks. Instead of treating the error recovery and

scheduling schemes in separate DLC and MAC layers, the scheduling scheme is

implemented in conjunction with a retransmission scheme. In particular, with the

aid of proper channel modeling, type I1 hybrid ARQ is chosen as the error

recovery scheme to combat fading effects while adaptive fair-queuing is adopted

as the scheduling scheme to achieve a fair and eacient resource allocation.

Specifically, the weight of a connection used in the fair-queuing algorithm

dynamically varies according to the current channel condition, as well as the type

of service. Various simulation results show that the proposed scheme can achieve

a high throughput with minimum delay and cell loss rate compared to

conventional algorithms.

The proposed Retransmission algorithms based on power priorities (RAPP) may

be used for the power control of the retransmission of request packets sent by

remote terminals to the base station (BS) in WATM networks. Simulation results

show that the throughput achieved is greater in this scheme.

a In this work, a new decentralised CAC is presented. The available bandwidth is

calculated at each node and updated regularly to be used by the CAC to route the

incoming call. The method proposed takes a quick decision on the connection

acceptance before the actual CAC takes the final decision. Mobility jnformation

is likely to be very usefirl in CAC for wireless ATM networks, as it is used to

estimate future resource requirements. The accuracy of resource estimation

which is essential to the CAC scheme was determined by the available mobility

information and the update time. Simulation results illustrate a considerable

improvement in the CAC responsiveness and in the overall network performance.

Under heavy traffic conditions, the proposed method performs very well. A

comparative analysis of the decentralised approach with the centralised approach

for different measures such as admission delays and throughput has been carried

out. The benefits of the proposed DCAC have been proved by the results.

Moreover9 the proposed CAC is simple enough that the admission decision can

be made in real time without much computational difficulties. Furthermore, the

handoff dropping and blocking probabilities are limited by the proposed CAC

scheme to a &urn level aImost independent of load conditions.